Fluid control intelligent adjusting method and system of micro-fluidic chip

Abstract

The invention relates to the technical field of fluid control, and discloses a fluid control intelligent adjusting method and system for a micro-fluidic chip, and the method comprises the steps: carrying out the data collection of a self-adaptive fluid controller, a micro-fluidic channel pressure sensor array, a temperature sensor array and a flow sensor array based on the micro-fluidic chip; obtaining real-time fluid parameter monitoring data; performing feature extraction to obtain a multi-dimensional fluid feature matrix; establishing a three-dimensional fluid model of the microfluidic channel, and solving to obtain flow field distribution data; inputting the flow field distribution data into a Kriging agent model, establishing a mapping model of channel width, depth, length and fluid control performance, and calculating to obtain an optimal geometric parameter combination; and performing multi-level calculation on the fluid pressure uniformity, the temperature distribution uniformity and the flow stability to obtain a hierarchical fluid parameter control strategy. According to the invention, the problem of multi-physics field coupling is effectively solved, the calculation complexity of the control system is reduced, and the execution efficiency of the control strategy is improved.

Description

Technical Field

[0001] The present invention relates to the field of fluid control technology, and in particular to a fluid control intelligent regulation method and system for a microfluidic chip. Background Art

[0002] Due to the complex fluid flow characteristics in microfluidic channels, traditional fluid control methods are difficult to meet the requirements of high precision and high stability, especially when facing dynamic changes under different working conditions, the control effect is poor.

[0003] At present, fluid control mainly relies on fixed control strategies, lacking the ability to monitor and dynamically adjust fluid parameters in real time. At the same time, there is a complex coupling relationship between the pressure distribution, temperature field and flow changes in the microfluidic channel, which makes it difficult for a single control method to ensure the accuracy and stability of fluid control. Summary of the invention

[0004] The present invention provides a fluid control intelligent regulation method and system for a microfluidic chip, which effectively solves the multi-physical field coupling problem, reduces the computational complexity of the control system, and improves the execution efficiency of the control strategy.

[0005] In a first aspect, the present invention provides a method for intelligently regulating fluid control of a microfluidic chip, the method comprising: Based on the microfluidic chip, data is collected from the adaptive fluid controller, microfluidic channel pressure sensor array, temperature sensor array and flow sensor array to obtain real-time fluid parameter monitoring data; Inputting the real-time fluid parameter monitoring data into a dynamic self-learning control network model for feature extraction to obtain a multi-dimensional fluid feature matrix; Establish a three-dimensional fluid model of the microfluidic channel within a set pressure range and temperature range, and solve to obtain the flow field distribution data; The flow field distribution data is input into a Kriging proxy model, a mapping model between channel width, depth, length and fluid control performance is established, and the optimal geometric parameter combination of the microfluidic channel is calculated by a non-dominated sorting genetic algorithm; According to the multi-dimensional fluid characteristic matrix and the optimal geometric parameter combination, multi-level calculations are performed on fluid pressure uniformity, temperature distribution uniformity and flow stability to obtain a hierarchical fluid parameter control strategy.

[0006] In a second aspect, the present invention provides a fluid control intelligent regulation system for a microfluidic chip, the fluid control intelligent regulation system for the microfluidic chip comprising: A data acquisition module is used to collect data from the adaptive fluid controller, the microfluidic channel pressure sensor array, the temperature sensor array and the flow sensor array based on the microfluidic chip to obtain real-time fluid parameter monitoring data; A feature extraction module, used for inputting the real-time fluid parameter monitoring data into a dynamic self-learning control network model for feature extraction to obtain a multi-dimensional fluid feature matrix; A solution module is used to establish a three-dimensional fluid model of the microfluidic channel within a set pressure range and temperature range, and solve to obtain flow field distribution data; Establishing a module for inputting the flow field distribution data into a Kriging proxy model, establishing a mapping model between channel width, depth, length and fluid control performance, and calculating the optimal geometric parameter combination of the microfluidic channel through a non-dominated sorting genetic algorithm; The calculation module is used to perform multi-level calculations on fluid pressure uniformity, temperature distribution uniformity and flow stability according to the multi-dimensional fluid characteristic matrix and the optimal geometric parameter combination, so as to obtain a hierarchical fluid parameter control strategy.

[0007] The third aspect of the present invention provides a computer device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the computer device executes the above-mentioned intelligent adjustment method for fluid control of the microfluidic chip.

[0008] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned intelligent adjustment method for fluid control of a microfluidic chip.

[0009] In the technical solution provided by the present invention, the present invention adopts a combination of an adaptive fluid controller and a multi-type sensor array to achieve high-precision real-time monitoring of fluid parameters in a microfluidic channel, thereby improving the accuracy of data acquisition; by constructing a dynamic self-learning control network model, adaptive extraction of fluid characteristics is achieved, and the system's adaptability to different working conditions is enhanced; the Kriging proxy model is innovatively combined with a non-dominated sorting genetic algorithm to establish an accurate prediction model for fluid control performance and optimize control parameters; the hierarchical fluid parameter control strategy proposed in the present invention achieves coordinated optimization of the three control layers of pressure, temperature, and flow through multi-level calculations, thereby improving control accuracy; the hierarchical optimization method is used for inter-layer coordination, which effectively solves the problem of multi-parameter coupling and ensures the stability of the control system. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0011] Figure 1 A schematic diagram of the steps of the fluid control intelligent regulation method of the microfluidic chip in an embodiment of the present invention; Figure 2 It is a structural schematic diagram of the fluid control intelligent regulation system of the microfluidic chip in an embodiment of the present invention; Figure 3 It is a schematic block diagram of the structure of a computer device in an embodiment of the present invention. DETAILED DESCRIPTION

[0012] Embodiments of the present invention provide a method and system for intelligent fluid control regulation of a microfluidic chip. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0013] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , an embodiment of the fluid control intelligent regulation method of the microfluidic chip in the embodiment of the present invention includes: Step S1, collecting data from the adaptive fluid controller, the microfluidic channel pressure sensor array, the temperature sensor array and the flow sensor array based on the microfluidic chip to obtain real-time fluid parameter monitoring data; It is understandable that the execution subject of the present invention may be a fluid control intelligent regulation system of a microfluidic chip, or a terminal or a server, which is not limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0014] Specifically, fine grid division is performed according to the structural characteristics of the microfluidic channel. The grid division is based on the geometric shape of the microfluidic channel, and the channel area is divided into multiple small units. The nodes of these units are designed as the layout positions of sensors to form a monitoring point distribution network composed of a pressure sensor array. This network structure can cover all key areas of the channel, ensuring that the pressure sensor array can collect accurate pressure data in real time. At the same time, based on the physical characteristics of the microfluidic channel, the key points of its inlet and outlet positions, channel corners and diversion nodes are calibrated, and the temperature sensor array is arranged at these calibration points. The temperature sensor array can collect real-time distribution data of the temperature in the microfluidic channel in a matrix form to form a temperature monitoring node matrix data set. At the same time, flow monitoring points are set at the feed end and discharge end of the microfluidic channel, and flow sensor arrays are arranged at these positions to ensure that the flow rate and flow rate changes of the fluid can be monitored and the fluid flow parameters can be obtained. In order to optimize the accuracy of the monitoring data, the fluid flow area is intelligently partitioned based on the reference fluid chamber of the adaptive fluid controller. The intelligent partitioning is centered on the pressure characteristics and divides the entire fluid area into multiple reference pressure monitoring point arrays. These reference pressure monitoring points work together with the distributed network of the pressure sensor array to calculate the real-time pressure difference by comparing the actual data of the pressure monitoring points with the reference pressure data in real time, forming a pressure gradient distribution sequence. This reflects the pressure distribution characteristics in the microfluidic channel. The pressure gradient distribution sequence, the temperature monitoring node matrix data, and the fluid flow parameters provided by the flow sensor are comprehensively analyzed and fused. Data fusion is achieved through a specific algorithm model, including weighted fusion, multi-source data calibration, and feature extraction of deep learning models. Through fusion processing, the data deviation generated by a single sensor is eliminated, and the accuracy and comprehensiveness of the monitoring data are improved. The fusion process combines the dynamic characteristics of the microfluidic chip to synchronize the different data types on the time axis to obtain a high-dimensional data set containing real-time fluid state information, namely real-time fluid parameter monitoring data.

[0015] Step S2, inputting the real-time fluid parameter monitoring data into the dynamic self-learning control network model for feature extraction to obtain a multi-dimensional fluid feature matrix; Specifically, the pressure gradient distribution sequence collected in real time is input into the adaptive fluid control module in the dynamic self-learning control network model for processing. The core of the adaptive fluid control module lies in the pressure feature extraction network and pressure parameter optimization network it contains. The pressure feature extraction network consists of three convolutional layers and two fully connected layers. Through the deep convolution operation of the pressure gradient distribution sequence, it can extract the complex pressure characteristics in the microfluidic channel, which reflect the subtle changes in the fluid pressure distribution in the channel. The multi-level feature extraction process of the convolution layer can effectively capture the local characteristics of the pressure gradient distribution, and at the same time, the global aggregation of features is realized through the fully connected layer to generate the pressure control feature sequence. In the pressure parameter optimization network, the extracted pressure features are processed by two fully connected layers and one output layer, and the pressure characteristics are optimized and parameterized to ensure that the obtained pressure control feature sequence has good interpretability and efficiency. At the same time, the temperature monitoring node matrix data in the real-time fluid parameter monitoring data is input into the spatiotemporal feature extraction module in the dynamic self-learning control network model for processing. This module consists of two parts: the temporal feature network and the spatial feature network, and realizes the efficient extraction of temperature distribution features through three-dimensional convolution operation. The temporal feature network adopts the long short-term memory network structure (LSTM), which can capture the dynamic change law of temperature data in the time dimension, and serialize these time characteristics to generate the temporal features of temperature change. At the same time, the spatial feature network captures the local characteristics of temperature distribution in the spatial dimension through three-dimensional convolution operation. This spatial feature reflects the change law of temperature distribution in different areas of the microfluidic chip. Through the synergy of the temporal feature network and the spatial feature network, the temperature monitoring node matrix data is efficiently deconstructed into a temperature distribution feature sequence, which not only contains multi-dimensional information of time and space, but also reflects the comprehensive characteristics of heat conduction and temperature fluctuation in the chip. The fluid flow parameters of the real-time fluid parameter monitoring data are input into the parameter self-adjustment module in the dynamic self-learning control network model. This module consists of a flow prediction network and a parameter correction network, and realizes high-precision modeling of flow change characteristics through a dynamic weight adjustment mechanism. The role of the flow prediction network is to model and predict fluid flow parameters, and use real-time monitoring data to generate prediction results for future flow change trends. At the same time, the parameter correction network dynamically adjusts the prediction results to correct the existing errors and generate a highly reliable flow change feature sequence to reflect the flow velocity fluctuation and flow stability changes of the fluid under different working conditions. The pressure control feature sequence, temperature distribution feature sequence and flow change feature sequence are fused through the dynamic self-learning control network model to generate a fused feature matrix. The matrix is ​​optimized through the time series reconstruction operation. The reconstruction process uses the continuity and dynamic change rules in the time dimension to integrate the fused feature matrix into a multi-dimensional fluid feature matrix.

[0016] Step S3, establishing a three-dimensional fluid model of the microfluidic channel within a set pressure range and temperature range, and solving to obtain flow field distribution data; Specifically, the pressure range and temperature range of the microfluidic channel are determined. These boundary conditions determine the physical constraints of the model and directly affect the flow characteristics of the fluid. On this basis, three-dimensional solid modeling is performed according to the geometric structure of the microfluidic channel. The model includes key areas of the channel, such as the feed end, the discharge end, the corners, and the diversion nodes. The modeling process uses CAD tools to ensure that the geometric shape can reflect the characteristics of the actual microfluidic chip. Hexahedral structured grid division is performed for the three-dimensional fluid model. Grid division needs to balance computational efficiency and accuracy. Conventional grids are used in general areas of the channel, while grid encryption methods are used in complex flow areas such as inlets and outlets, corners, and diversion nodes. By refining the grids of these key areas, the complexity of local fluid behavior can be effectively captured and the accuracy of the calculation can be improved. After the grid division is completed, the mass conservation equation, momentum conservation equation, and energy conservation equation are respectively applied to the computational domain grid, and the finite volume method is used to discretize these conservation equations. The finite volume method adapts to numerical calculations by converting continuous equations into discrete forms. Specifically, the second-order upwind scheme is used for spatial discretization to improve the stability and accuracy of the calculation. The control equations formed after discretization are the basis of fluid mechanics calculations and can describe the physical changes of the fluid in the channel. The discrete control equations are solved based on the SIMPLE algorithm. The SIMPLE algorithm is an iterative algorithm that can effectively deal with the pressure-velocity coupling problem in fluid calculations. By initially guessing the flow field value, the pressure and velocity fields are gradually corrected to obtain the initial flow field solution. In order to optimize the accuracy and efficiency of the calculation, the number of grids is adjusted based on the preliminary flow field solution, that is, the grid density is increased in the area where the fluid behavior changes drastically, and the number of grids is appropriately reduced in the area where the flow changes slowly to generate the optimal grid distribution. The initial value of the flow field is iteratively solved on the optimal grid distribution. The steady-state flow field solution is obtained by stepwise approximation. The steady-state flow field solution is the final fluid state of the system under given boundary conditions, reflecting the stable distribution of the pressure field, velocity field and temperature field. From this steady-state solution, detailed data of the fluid are extracted, including pressure distribution, temperature gradient and flow velocity distribution. These data together constitute the flow field distribution data, which describes the fluid behavior in the microfluidic channel.

[0017] Step S4, inputting the flow field distribution data into the Kriging proxy model, establishing a mapping model between channel width, depth, length and fluid control performance, and calculating the optimal geometric parameter combination of the microfluidic channel through a non-dominated sorting genetic algorithm; Specifically, the flow field distribution data are used as sample points. These data contain the distribution information of the pressure field, temperature field and flow field in the microfluidic channel, which can fully reflect the dynamic characteristics of the fluid. On this basis, the Latin hypercube sampling method is used to generate the training sample space for the geometric parameter range of the microfluidic channel, including the width range, depth range and length range. Latin hypercube sampling can evenly distribute the sampling points in the high-dimensional space to ensure that the entire geometric parameter range is covered, while avoiding the overlap and clustering of the sampling points, providing a diverse and efficient sample set for the training of subsequent models. Based on the generated training sample space, a Kriging proxy model is constructed, which establishes the mapping relationship between geometric parameters and fluid properties through the interpolation method. The core of the Kriging proxy model is to use the Gaussian correlation function as the kernel function, set the correlation parameter θ to 0.1, and set the order of the regression function to 2, thereby improving the model's ability to fit nonlinear features. By calculating the correlation matrix of the geometric parameters, the mutual influence relationship between different parameters can be captured. On this basis, the maximum likelihood estimation function is used to optimize the hyperparameters and determine the optimal interpolation weight of the model. The optimization result of the interpolation weight directly determines the prediction accuracy of the Kriging proxy model. By substituting it into the model, the mapping model between the channel width, depth and length and the fluid control performance index is constructed. The established mapping model is used to optimize the solution through the non-dominated sorting genetic algorithm. The non-dominated sorting genetic algorithm is a multi-objective optimization method that can search for the optimal combination of geometric parameters while meeting multiple performance requirements. In the optimization process, the initial population is generated, the fitness of the individuals in the population is evaluated based on the evolutionary process, and the Pareto optimal solution is calculated according to the multi-objective optimization standard. In order to ensure the comprehensiveness and stability of the optimization process, the individuals are sorted by the crowding distance, and the individuals with a non-dominated level of 1 are selected as the elite population to form the Pareto front. The solution set in the Pareto front represents the combination of geometric parameters that meets the multi-objective optimization conditions. These combinations achieve the optimal trade-off between performance indicators such as pressure uniformity, temperature uniformity and flow stability. In order to select the optimal solution from the Pareto front, the solution set is normalized. By mapping different performance indicators to a unified scale, it is convenient to compare multi-objective performance. After normalization, candidate parameter combinations are obtained, all of which meet the requirements of microfluidic channel optimization design. In order to select the optimal parameter combination, sensitivity analysis is performed to analyze the influence of each geometric parameter on fluid control performance and comprehensively evaluate the stability and reliability of the parameter combination. Based on the comprehensive evaluation value, the optimal geometric parameter combination is selected as the design scheme of the microfluidic channel.

[0018] The regression function matrix is ​​constructed on the basis of the training sample space. In order to enhance the model's ability to express the characteristics of the sample points, the second-order polynomial basis function is used for expansion. The second-order polynomial basis function expansion can capture the nonlinear characteristics of the input sample points. By analytically solving the regression matrix, the coefficients of the regression model are obtained, reflecting the basic influence of the input geometric parameters on the response target. The coefficients of the regression model are substituted into the Gaussian correlation function, and the spatial correlation between the sample points is evaluated by the calculation method of the kernel function. The Gaussian correlation function takes the distance between the sample points as a variable. By setting the correlation parameter θ to 0.1, it is ensured that the kernel function has an appropriate attenuation characteristic for the correlation of the sample points. This attenuation characteristic can effectively characterize the variation law of the input geometric parameters in the spatial distribution, and generate the correlation function matrix. The correlation function matrix is ​​the core description of the spatial dependence between the sample points. In order to simplify the processing of high-dimensional data, the matrix is ​​subjected to feature dimensionality reduction decomposition. Dimensionality reduction maps the original high-dimensional data to a low-dimensional feature space by performing eigenvalue decomposition on the correlation function matrix, while retaining the main change information of the data. The obtained reduced-dimensional feature space can significantly reduce the computational complexity while ensuring information integrity. The maximum likelihood estimation function is constructed based on the reduced-dimensional feature space. The maximum likelihood estimation function aims to find the optimal value of the hyperparameters through optimization, and its goal is to minimize the prediction error of the Kriging model. To facilitate numerical calculation and optimization, the function uses a logarithmic transformation to convert complex nonlinear optimization problems into a form that is more suitable for numerical solutions. By inputting the optimization objective function into the pattern search algorithm, the optimal hyperparameters can be found efficiently. The pattern search algorithm is a global optimization method that does not require gradient information and can find the global optimal solution in complex multi-peak optimization problems, thereby ensuring that the obtained hyperparameters are reliable and stable. After obtaining the optimal hyperparameters, the accuracy of the model is verified. The model is trained and tested multiple times through the cross-validation method to evaluate the performance of the Kriging proxy model on different sample sets. Cross-validation can effectively avoid the problem of model overfitting, and at the same time generate model accuracy evaluation indicators, such as mean square error (MSE), coefficient of determination ( ) and so on. These indicators provide a quantitative basis for the modification of the model. On this basis, the Kriging proxy model is modified according to the accuracy evaluation index to ensure that its prediction ability meets the requirements of practical applications. The optimal interpolation weights of the Kriging model are calculated by the generalized least squares method. The generalized least squares method combines the coefficients of the regression model, the correlation function matrix and the optimized hyperparameters to improve the accuracy of the interpolation results while ensuring the stability of the model. The final interpolation weights are the core of the Kriging proxy model, which can make high-precision response predictions for new input samples, thereby providing accurate performance mapping for the geometric parameter optimization of the microfluidic chip.

[0019] Step S5: Based on the multi-dimensional fluid characteristic matrix and the optimal geometric parameter combination, multi-level calculations are performed on the fluid pressure uniformity, temperature distribution uniformity and flow stability to obtain a hierarchical fluid parameter control strategy.

[0020] Specifically, the quantitative relationship between these fluid characteristics and geometric parameters is revealed by performing correlation analysis on the pressure control feature sequence, temperature distribution feature sequence and flow change feature sequence in the multidimensional fluid feature matrix and the optimal geometric parameter combination. The correlation analysis maps the relationship between the feature matrix and the geometric parameters into a multidimensional space based on statistical and machine learning methods to form a feature-parameter mapping relationship, which reflects the contribution and sensitivity of the geometric parameters to the fluid control performance. The feature-parameter mapping relationship is input into the control strategy generation module to construct an initial hierarchical control framework including the pressure control layer, the temperature control layer and the flow control layer. The complex fluid control problem is decomposed into sub-problems that can be optimized independently. By processing the pressure, temperature and flow control in layers, the complexity of optimization is significantly simplified, while the pertinence and efficiency of each layer of control are improved. In the pressure control layer, the pressure data is dynamically programmed based on the initial hierarchical control framework. Dynamic programming decomposes the overall optimization problem into sub-problems of multiple stages and solves them through a recursive method, so that the decision of each stage is optimal, thereby ensuring that the overall pressure uniformity reaches the optimal level. Thus, the pressure uniformity control parameters are generated as the optimization output of the pressure layer. In the temperature control layer, model predictive control (MPC) is used to optimize the temperature data based on the initial hierarchical control framework. MPC is based on the real-time prediction of temperature distribution. By solving the rolling optimization problem, the control parameters that can maximize the uniformity of temperature distribution are calculated. Its core lies in real-time prediction of temperature change trend and dynamic adjustment of control strategy according to the prediction results, so as to effectively deal with the temperature fluctuation problem in the microfluidic system and generate temperature distribution uniformity control parameters. At the same time, in the flow control layer, robust control calculation is used for flow data based on the initial hierarchical control framework. The goal of robust control is to deal with the uncertainty of flow change and ensure the flow stability of the fluid under different operating conditions. Through the robust control method, the control strategy can be optimized to minimize flow fluctuations and generate flow stability control parameters in the presence of interference and modeling errors. The pressure uniformity control parameters, temperature distribution uniformity control parameters and flow stability control parameters are input into the hierarchical coordinator. The role of the hierarchical coordinator is to achieve coordination and consistency between layers through hierarchical optimization. Hierarchical optimization is based on the dependencies between layers, and the optimization results are gradually passed down from the top-level decision, while the feedback information from the lower layer is passed to the upper layer to ensure the coordination and optimality of the overall system performance. In the process of hierarchical optimization, the output parameters of each control layer are dynamically adjusted to meet the global optimization goal, and finally a hierarchical fluid parameter control strategy is generated.

[0021] In the embodiments of the present invention, the present invention adopts a combination of an adaptive fluid controller and a multi-type sensor array to achieve high-precision real-time monitoring of fluid parameters in a microfluidic channel, thereby improving the accuracy of data acquisition; by constructing a dynamic self-learning control network model, adaptive extraction of fluid characteristics is achieved, and the system's adaptability to different working conditions is enhanced; the Kriging proxy model is innovatively combined with a non-dominated sorting genetic algorithm to establish an accurate prediction model for fluid control performance and optimize control parameters; the hierarchical fluid parameter control strategy proposed in the present invention achieves coordinated optimization of the three control layers of pressure, temperature, and flow through multi-level calculations, thereby improving control accuracy; the hierarchical optimization method is used for inter-layer coordination, which effectively solves the problem of multi-parameter coupling and ensures the stability of the control system.

[0022] In a specific embodiment, the process of executing step S1 may specifically include the following steps: The microfluidic channel is gridded, and the pressure sensor array is arranged at the grid nodes to obtain a pressure monitoring point distribution network; the key points of the microfluidic channel inlet and outlet, corners and diversion nodes are calibrated, and the temperature sensor array is arranged at the calibration point position to collect the temperature monitoring node matrix data; the flow monitoring points are set at the feed end and discharge end of the microfluidic channel, and the flow sensor array is arranged at the monitoring point to collect the fluid flow parameters; Based on the adaptive fluid controller, the reference fluid chamber is intelligently partitioned to obtain a reference pressure monitoring point array, and the pressure data of the pressure monitoring point distribution network and the pressure data of the reference pressure monitoring point array are subjected to real-time difference calculation to obtain a pressure gradient distribution sequence; Based on the microfluidic chip, the pressure gradient distribution sequence, temperature monitoring node matrix data and fluid flow parameters are fused to obtain real-time fluid parameter monitoring data.

[0023] Specifically, the microfluidic channel is meshed. The internal geometric structure of the channel is divided into small units to provide a more detailed description of the fluid behavior. In this process, the pressure sensor array is arranged on each grid node, which is the key position after meshing. Assume that the geometric area of ​​the channel is divided into The center position of each grid unit is defined as the grid node position. , the pressure sensor will record the local pressure value , forming a three-dimensional pressure monitoring point distribution network. This network can capture the global pressure distribution in the channel and reflect the local pressure gradient changes through the difference between nodes. After completing the grid division and pressure sensor layout, the key positions of the microfluidic channel are calibrated, including the inlet and outlet of the channel, corners, and diversion nodes. These locations are areas where the fluid behavior changes significantly, such as sudden changes in pressure, changes in flow velocity direction, or changes in flow distribution. When calibrating these key points ( ), place the temperature sensor at the calibration point to collect temperature data These temperature data can reflect the heat conduction characteristics and temperature change rules of the fluid in the key area, forming a temperature monitoring node matrix data set. At the same time, flow monitoring points are set at the feed end and the discharge end of the microfluidic channel. Assuming that there is Monitoring points, discharge end has monitoring points, and the flow data of each monitoring point is recorded as and By arranging a flow sensor array and recording real-time flow parameters, the flow distribution and stability of the channel can be accurately grasped. For a single monitoring point, its flow data is obtained by integration: ; in, is the flow velocity vector, is the normal vector of the section at the monitoring point, is the cross-sectional area of ​​the monitoring point. Similarly, the flow data at the discharge end is calculated in the same way. Based on the reference fluid chamber of the adaptive fluid controller, the channel is intelligently partitioned. The reference area is divided according to the fluid control requirements to generate a reference pressure monitoring point array. Assume that the reference fluid chamber is divided into sub-areas, and the center of each sub-area is defined as the reference pressure monitoring point The base pressure is By calculating the data of the pressure monitoring point distribution network The difference between the pressure value and the reference pressure value gives the pressure gradient distribution sequence: ; in, Indicates The pressure gradient at each grid node. The gradient sequence reflects the direction and intensity of the pressure change in the channel.

[0024] In a specific embodiment, the process of executing step S2 may specifically include the following steps: The pressure gradient distribution sequence of the real-time fluid parameter monitoring data is input into the adaptive fluid control module in the dynamic self-learning control network model for processing to obtain a pressure control feature sequence. The adaptive fluid control module includes a pressure feature extraction network and a pressure parameter optimization network; the pressure feature extraction network includes three convolutional layers and two fully connected layers, and the pressure parameter optimization network includes two fully connected layers and one output layer; The temperature monitoring node matrix data of the real-time fluid parameter monitoring data is input into the spatiotemporal feature extraction module in the dynamic self-learning control network model for three-dimensional convolution operation to obtain the temperature distribution feature sequence. The spatiotemporal feature extraction module includes a time series feature network and a space feature network. The time series feature network adopts a long short-term memory network structure, and the space feature network adopts a three-dimensional convolution structure. The fluid flow parameters of the real-time fluid parameter monitoring data are input into the parameter self-adjustment module in the dynamic self-learning control network model for dynamic weight adjustment to obtain a flow change characteristic sequence, and the parameter self-adjustment module includes a flow prediction network and a parameter correction network; The pressure control feature sequence, the temperature distribution feature sequence and the flow change feature sequence are fused to obtain a fused feature matrix, and the fused feature matrix is ​​time-reconstructed to obtain a multi-dimensional fluid feature matrix.

[0025] Specifically, the pressure gradient distribution sequence in the real-time fluid parameter monitoring data is input into the adaptive fluid control module in the dynamic self-learning control network model for processing. This module contains a pressure feature extraction network and a pressure parameter optimization network, which extracts key features from the input pressure gradient distribution sequence and optimizes these features to generate a pressure control feature sequence. In the pressure feature extraction network, the data is processed by three convolutional layers and two fully connected layers. Assume that the input pressure gradient distribution sequence is ,in represents the grid node index, Represents the time dimension. The convolution operation is performed through the weighted convolution kernel Extract local spatial features, the formula is: ; in, It is The output of the convolutional layer, is the activation function, is the bias term, is the local index of the convolution kernel. After three layers of convolution, the feature data is input into two layers of fully connected layers for global feature aggregation. The output of the fully connected layer is expressed as: ; in, is the weight matrix of the fully connected layer, is the feature input of the previous layer, is a bias term. The pressure feature extraction network outputs a high-dimensional pressure feature vector The pressure feature vector is input into the pressure parameter optimization network, which contains two fully connected layers and one output layer. Through weight optimization and nonlinear mapping, the pressure control feature sequence is obtained. , this sequence can reflect the key change characteristics of pressure in the microfluidic channel. At the same time, the temperature monitoring node matrix data in the real-time fluid parameter monitoring data is input into the spatiotemporal feature extraction module of the dynamic self-learning control network model. This module contains a temporal feature network and a spatial feature network, which respectively deal with dynamic changes and spatial distribution characteristics in the time dimension. In the spatial feature network, a three-dimensional convolution structure is used to extract the spatial distribution characteristics of temperature. The convolution calculation is similar to the pressure feature extraction network, but in the three-dimensional convolution, each voxel of the input data is convolved with a three-dimensional convolution kernel to generate a spatial feature map. In the temporal feature network, a long short-term memory network (LSTM) structure is used to model the features in the time dimension. Assume that the temperature matrix of the time series is , the calculation formula of LSTM is: ; in, It's time The hidden state of is the weight matrix of LSTM, is the bias term. Through the processing of this network, the temperature distribution feature sequence is generated , which can capture the dynamic characteristics of temperature changes. The fluid flow parameters are input into the parameter self-adjustment module, which contains the flow prediction network and the parameter correction network. The flow prediction network is based on historical flow data. Build a forecasting model to generate estimates of future traffic changes: ; in, is the weight matrix of the prediction network, is the bias term. The parameter correction network dynamically adjusts the weight of the predicted value to generate a traffic change feature sequence , to reflect the stability of fluid flow. The pressure control feature sequence, temperature distribution feature sequence and flow change feature sequence are fused to generate a fusion feature matrix . Feature fusion is achieved through weighted summation: ; in, are the weight coefficients of pressure, temperature and flow characteristics respectively. The fusion feature matrix generates a multi-dimensional fluid feature matrix through time series reconstruction operation , the features at different time steps are fused into a unified high-dimensional space, providing a comprehensive description of the fluid state.

[0026] In a specific embodiment, the process of executing step S3 may specifically include the following steps: Determine the pressure range and temperature range of the microfluidic channel, and perform three-dimensional solid modeling on the microfluidic channel to obtain a three-dimensional fluid model including a feed end, a discharge end, a corner, and a flow diversion node; The three-dimensional fluid model is divided into hexahedral structured grids, and the grids at the inlet and outlet, corners and flow diversion nodes are encrypted to obtain the computational domain grid; The mass conservation equation, momentum conservation equation and energy conservation equation are discretized on the computational domain grid using the finite volume method, and the second-order upwind scheme is used for spatial discretization to obtain the discrete control equations. The discrete control equations are solved based on the SIMPLE algorithm to obtain the initial value of the flow field, and the number of grids is encrypted and coarsened based on the initial value of the flow field to obtain the optimal grid distribution; The initial value of the flow field is iteratively solved on the optimal grid distribution to obtain a steady-state flow field solution, and based on the steady-state flow field solution, the fluid pressure field, temperature field and velocity field data are extracted to obtain the flow field distribution data.

[0027] Specifically, determine the pressure range and temperature range of the microfluidic channel. These boundary conditions determine the physical properties of the fluid and the constraints in the solution process. Set the pressure range to , the temperature range is ,in and represent the lowest and highest pressures in the channel, respectively. and Represent the minimum and maximum temperatures respectively. Three-dimensional solid modeling is performed based on the actual geometric structure of the microfluidic channel. The geometric model includes key structures such as the feed end, discharge end, corners and diversion nodes of the channel to accurately describe the flow path and dynamic characteristics of the fluid. The three-dimensional geometric model is divided into hexahedral structured grids. The continuous computational domain is discretized into multiple finite elements so that numerical calculations can be performed on these elements. Assume that the number of elements after meshing is ,in Respectively represent the grid , , The number of distributions in the direction. In order to improve the calculation accuracy, the key areas of the geometric structure (such as inlets and outlets, corners and flow diversion nodes) are meshed. The mesh size after encryption is smaller, so as to more finely capture the changing characteristics of the local flow. For example, at the feed end, the mesh size The mesh size is set to 1 / 5 of the overall mesh size to ensure the accuracy of the inlet and outlet flows. The mass conservation equations, momentum conservation equations, and energy conservation equations are discretized onto the computational domain mesh. For any mesh element , the mass conservation equation is expressed as: ; in, is the fluid density, is the velocity vector. The momentum conservation equation is expressed as: ; in, For pressure, is the dynamic viscosity. The energy conservation equation is: ; in, is the specific internal energy, is the heat flux, is the viscous dissipation term. The above equations are discretized by the finite volume method, and the second-order upwind scheme is used for spatial discretization to improve the calculation accuracy and stability. After discretization, the mass conservation equation is The discretization form of a grid unit is: ; in, is the normal area vector of the mesh surface, and are the density and velocity values ​​on the grid surface respectively. Based on the above discrete equations, the SIMPLE algorithm is used to solve them. The SIMPLE algorithm solves the coupling problem of nonlinear equations by step-by-step correction of pressure and velocity fields. Assuming an initial flow field , calculate the speed correction term and pressure correction term , and then update the pressure and velocity fields: ; in, and is the relaxation factor, Represents the number of iterations. After multiple iterations, the initial value of the flow field is finally obtained. Based on the initial value of the flow field, the number of grids is encrypted and coarsened to form the optimal grid distribution. The principle of grid optimization is to increase the grid density in areas where the fluid behavior changes dramatically (such as high gradient areas) and reduce the grid density in areas where the changes are gentle. For example, in the fluid velocity gradient Above a certain threshold In the area of ​​, the grid is encrypted to half the original grid size to improve the calculation accuracy. On the optimal grid distribution, the initial flow field is used as the starting point and the solution is iterated until convergence to obtain the steady-state flow field solution. The steady-state flow field solution contains the pressure field of the fluid , Temperature field and velocity field .

[0028] In a specific embodiment, the process of executing step S4 may specifically include the following steps: The flow field distribution data is used as sample points, and Latin hypercube sampling is performed on the width range, depth range and length range of the microfluidic channel to obtain the training sample space; A Kriging proxy model is established based on the training sample space. The Gaussian correlation function is used as the kernel function. The correlation parameter θ is set to 0.1, and the order of the regression function is set to 2. The correlation matrix of the geometric parameters is obtained. The maximum likelihood estimation function is constructed based on the correlation matrix. The hyperparameters are optimized to obtain the optimal interpolation weights of the Kriging model. Substitute the optimal interpolation weights into the Kriging proxy model to establish a mapping model between channel width, depth, length and pressure uniformity, temperature uniformity, and flow stability; Based on the mapping model, the non-dominated sorting genetic algorithm is optimized to obtain the evolutionary population, and the Pareto optimal solution is calculated based on the evolutionary population. The crowding distance is used to sort individuals, and individuals with a non-dominated level of 1 are selected to form an elite population to obtain the Pareto frontier. The solution set in the Pareto front is normalized according to pressure uniformity, temperature uniformity and flow stability to obtain candidate parameter combinations; A sensitivity analysis was performed on the candidate parameter combinations, and the parameter combination with the best comprehensive evaluation value was selected as the optimal geometric parameter combination of the microfluidic channel.

[0029] Specifically, the flow field distribution data is used as sample points, combined with the geometric parameter range of the microfluidic channel, including the width range , Depth Range and length range , the Latin hypercube sampling method is used to generate the training sample space. Latin hypercube sampling is an efficient multidimensional space sampling method that ensures the uniform distribution of samples in the geometric parameter space by evenly dividing the parameter range and randomly arranging the sample point positions in each dimension. Assuming the number of samples is , for each sample point , the width, depth and length are generated according to the following formulas: ; ; ; in, is the random permutation index in Latin hypercube sampling. In the generated training sample space, the flow field distribution data is used as the response variable to establish the Kriging proxy model. The Kriging model uses the Gaussian correlation function as the kernel function, and its formula is: ; in, and For any two sets of geometric parameters, is the relevant parameter, is the normalized scaling factor for the parameter range. The regression function takes the form of a second-order polynomial: ; in, is the regression coefficient, which represents the linear and quadratic nonlinear effects of the geometric parameters on the target response. Combining the Gaussian correlation function and the regression function, the correlation matrix of the sample points is calculated , whose elements are . The optimization objective function is constructed by maximum likelihood estimation: ; in, is the determinant of the correlation matrix, is the response variable vector, is the number of samples. By optimizing Maximize the objective function to obtain the optimal hyperparameters and interpolation weights. Substitute the optimal interpolation weights into the Kriging model to establish a mapping model between geometric parameters and pressure uniformity, temperature uniformity, and flow stability: ; in, is the predicted value, is the correlation function vector, is the regression matrix, is the regression coefficient. The non-dominated sorting genetic algorithm (NSGA-II) is optimized based on the mapping model. After the population is initialized, the fitness of each individual is calculated according to the mapping model. The evolving population is generated through selection, crossover and mutation operations. The Pareto optimal solution set in the population is calculated, and its goal is to maximize the pressure uniformity, temperature uniformity and flow stability. In the screening process of the Pareto optimal solution, the crowding distance is used for sorting, and individuals with a non-dominated level of 1 are preferentially selected to form an elite population to obtain the Pareto front. The solution set in the Pareto front is normalized according to the three performance indicators. The normalization formula is: ; in, is the normalized value, and are the minimum and maximum values ​​of the corresponding performance indicators respectively. After normalization, the candidate parameter combination is obtained. The sensitivity analysis of the candidate parameter combination is performed to calculate the sensitivity index of each geometric parameter to the performance indicator. , the formula is: ; in, represents the variance of the predicted value when a certain parameter is fixed. Represents the variance of the overall prediction value. The sensitivity analysis results are comprehensively evaluated, and the parameter combination with the best comprehensive evaluation value is selected as the optimal geometric parameter combination of the microfluidic channel.

[0030] In a specific embodiment, the execution step establishes a Kriging proxy model based on the training sample space, uses the Gaussian correlation function as the kernel function, sets the correlation parameter θ to 0.1, and the regression function order to 2, obtains the correlation matrix of the geometric parameters, and constructs the maximum likelihood estimation function based on the correlation matrix, optimizes the hyperparameters, and obtains the optimal interpolation weight of the Kriging model. The process can specifically include the following steps: The regression function matrix is ​​constructed based on the training sample space, and the regression model coefficients are obtained by using the second-order polynomial basis function expansion; Substitute the regression model coefficients into the Gaussian correlation function, calculate the spatial correlation between the sample points, obtain the correlation function matrix, and perform feature dimensionality reduction decomposition on the correlation function matrix to obtain the reduced-dimensional feature space; The maximum likelihood estimation function is constructed based on the reduced-dimensional feature space, and the optimization objective function is obtained by logarithmic transformation. The optimization objective function is then input into the pattern search algorithm to obtain the optimal hyperparameters. The optimal hyperparameters are cross-validated to obtain the model accuracy evaluation index, and the Kriging proxy model is modified according to the model accuracy evaluation index. The optimal interpolation weights of the Kriging model are obtained by generalized least squares calculation.

[0031] Specifically, the regression function matrix is ​​constructed for the data in the training sample space. Assume that the training sample space contains sample points, each sample point is a The target response value is The regression function is expanded using a second-order polynomial basis function to represent the basic relationship between sample points. The specific form is: ; in, is a constant term, and are the regression coefficients of the first and second terms, respectively, indicating the linear and nonlinear effects of the sample points on each dimension. And the corresponding output , construct the regression matrix and the response vector : ; ; Regression model coefficient vector Calculated by the least squares method: ; Substitute the calculated regression coefficient into the Gaussian correlation function to construct the spatial correlation matrix between sample points. The form of the Gaussian correlation function is: ; in, Represents sample points and The correlation between It is The correlation parameters of the dimensions are used to control the decay rate of the correlation function. By calculating the correlation of all sample point pairs, the correlation function matrix is ​​constructed. : ; Perform feature dimensionality reduction decomposition on the correlation function matrix to obtain the reduced dimensionality feature space. The eigenvalue of and the eigenvector , the high-dimensional correlation is reduced to a linear combination: ; in, is the number of features after dimensionality reduction, select the main eigenvalue The corresponding feature vectors construct the reduced-dimensional feature space, thereby reducing the computational complexity. Based on the reduced-dimensional feature space, the maximum likelihood estimation function is constructed, and the logarithmic transformation form is used to transform the optimization objective function into: ; in, represents the determinant of the correlation matrix, is the weighted sum of squares of the prediction errors. Optimized by the pattern search algorithm , get the optimal hyperparameters The optimal hyperparameters were cross-validated, and the prediction accuracy of the model was evaluated by dividing the training set and the test set. Evaluation indicators such as mean square error (MSE) and determination coefficient ( ): ; ; in, is the model prediction value, is the true value, is the mean of the true value. The Kriging model is modified according to the evaluation index, and the optimal interpolation weight is calculated by the generalized least squares method. : ; The final constructed Kriging model accurately predicts the target response value.

[0032] In a specific embodiment, the process of executing step S5 may specifically include the following steps: The pressure control feature sequence, temperature distribution feature sequence and flow rate change feature sequence in the multi-dimensional fluid feature matrix are respectively correlated with the optimal geometric parameter combination to obtain the feature-parameter mapping relationship; The feature-parameter mapping relationship is input into the control strategy generation module to construct the initial hierarchical control framework of the pressure control layer, temperature control layer and flow control layer; Based on the initial hierarchical control framework, the pressure control layer data is dynamically programmed to obtain the pressure uniformity control parameters; Based on the initial hierarchical control framework, the temperature control layer data is model predicted and controlled to obtain the temperature distribution uniformity control parameters; Based on the initial hierarchical control framework, robust control calculation is performed on the flow control layer data to obtain the flow stability control parameters; The pressure uniformity control parameters, temperature distribution uniformity control parameters and flow stability control parameters are input into the hierarchical coordinator, and hierarchical optimization is used to perform inter-layer coordination to obtain the hierarchical fluid parameter control strategy.

[0033] Specifically, the pressure control feature sequence, temperature distribution feature sequence, and flow change feature sequence are associated with the optimal geometric parameter combination to construct a feature-parameter mapping relationship. Assume that the multidimensional fluid feature matrix is ,in represents the pressure control characteristic sequence, represents the characteristic sequence of temperature distribution, represents the characteristic sequence of flow change, is the number of samples. The optimal geometric parameter combination is recorded as ,in Respectively represent the width, depth and length of the channel. Through association analysis, a multivariate linear regression model or a nonlinear kernel method is used to establish a mapping relationship between each feature sequence and the geometric parameter combination, for example: ; ; ; in, is a constant term, are regression coefficients, is the error term. The regression coefficient is optimized by the least squares method to construct a complete feature-parameter mapping relationship. These mapping relationships are input into the control strategy generation module to construct the initial hierarchical control framework of the pressure control layer, temperature control layer, and flow control layer. Each control layer optimizes specific performance targets to ensure pressure uniformity, temperature distribution uniformity, and flow stability. Taking the pressure control layer as an example, its initial input is the feature-parameter mapping relationship and the pressure control feature sequence The pressure control layer data is optimized by dynamic programming. Dynamic programming decomposes the global optimization problem into a series of phased sub-problems. Each phase is solved recursively to ensure the overall optimality. Assume that the pressure uniformity objective function is: ; in, is the target pressure value. The optimization update formula of dynamic programming is: ; in, It is a stage The loss function is is the control variable. The final output pressure uniformity control parameter Similarly, in the temperature control layer, the model predictive control method is used to optimize the temperature distribution characteristic sequence. Model predictive control performs rolling optimization on temperature data based on the prediction model, and its objective function is: ; in, is the target temperature value, is the weight factor of the control increment. By optimizing the control input within the rolling window , generate temperature distribution uniformity control parameters. In the flow control layer, the robust control method is used to optimize the flow change characteristic sequence to deal with the uncertainty and interference in the system. The flow stability objective function is: ; in, is the target flow value, is the robust control weight coefficient, For the By solving the robust control problem, the flow stability control parameters are output. The optimization results of the three control layers, namely the pressure uniformity control parameters , Temperature distribution uniformity control parameters and flow stability control parameters , input the hierarchical coordinator for hierarchical optimization. The goal of hierarchical optimization is to coordinate the control strategies of each layer so that the global performance target is optimized. The objective function of hierarchical optimization is: ; in, is the weight coefficient of each layer. Through hierarchical optimization, the control parameters of each layer are adjusted to coordinate the optimization results of pressure, temperature and flow rate in the global scope, and the hierarchical fluid parameter control strategy is obtained.

[0034] Among them, the pressure uniformity control parameters, temperature distribution uniformity control parameters and flow stability control parameters are input into the hierarchical coordinator, and the inter-layer coordination is carried out by hierarchical optimization to obtain the hierarchical fluid parameter control strategy, including: performing hierarchical coupling degree analysis on the pressure uniformity control parameters, temperature distribution uniformity control parameters and flow stability control parameters to obtain the parameter coupling relationship matrix; inputting the parameter coupling relationship matrix into the hierarchical coordinator, and using the hierarchical analysis method to calculate the weight coefficient of each control layer to obtain the inter-layer weight distribution scheme; constructing the hierarchical optimization objective function based on the inter-layer weight distribution scheme, setting the pressure layer priority to 1, the temperature layer priority to 2, and the flow layer priority to 3, and obtaining the hierarchical optimization sequence; performing the first-order optimization calculation on the pressure layer in the hierarchical optimization sequence, and setting the pressure uniformity constraint condition to ±0.5MPa , and obtain the optimal control value of the pressure layer; substitute the optimal control value of the pressure layer into the temperature layer for second-order optimization calculation, set the temperature uniformity constraint to ±1℃, and obtain the optimal control value of the temperature layer; perform third-order optimization calculation on the flow layer based on the optimal control value of the pressure layer and the optimal control value of the temperature layer, set the flow stability constraint to ±0.1ml / s, and obtain the optimal control value of the flow layer; verify the coordination of the optimal control value of the pressure layer, the optimal control value of the temperature layer, and the optimal control value of the flow layer, and use the cross-validation method to calculate the degree of mutual influence of the control parameters of each layer to obtain a coordinated control sequence; organize the coordinated control sequence in time sequence according to the control period, set the pressure control period to 0.001s, the temperature control period to 0.01s, and the flow control period to 0.1s, and obtain a hierarchical fluid parameter control strategy.

[0035] The above describes the fluid control intelligent regulation method of the microfluidic chip in the embodiment of the present invention. The following describes the fluid control intelligent regulation system of the microfluidic chip in the embodiment of the present invention. Figure 2 In one embodiment of the present invention, an intelligent fluid control system of a microfluidic chip includes: A data acquisition module is used to collect data from the adaptive fluid controller, the microfluidic channel pressure sensor array, the temperature sensor array and the flow sensor array based on the microfluidic chip to obtain real-time fluid parameter monitoring data; A feature extraction module is used to input the real-time fluid parameter monitoring data into the dynamic self-learning control network model for feature extraction to obtain a multi-dimensional fluid feature matrix; A solution module is used to establish a three-dimensional fluid model of the microfluidic channel within a set pressure range and temperature range, and solve to obtain flow field distribution data; Establish a module for inputting flow field distribution data into the Kriging proxy model, establish a mapping model between channel width, depth, length and fluid control performance, and calculate the optimal geometric parameter combination of the microfluidic channel through the non-dominated sorting genetic algorithm; The calculation module is used to perform multi-level calculations on fluid pressure uniformity, temperature distribution uniformity and flow stability according to the multi-dimensional fluid characteristic matrix and the optimal geometric parameter combination, and obtain a hierarchical fluid parameter control strategy.

[0036] Through the coordinated cooperation of the above-mentioned components, the present invention adopts a combination of an adaptive fluid controller and a multi-type sensor array to achieve high-precision real-time monitoring of fluid parameters in the microfluidic channel, thereby improving the accuracy of data acquisition; by constructing a dynamic self-learning control network model, adaptive extraction of fluid characteristics is achieved, and the system's adaptability to different working conditions is enhanced; the Kriging proxy model is innovatively combined with a non-dominated sorting genetic algorithm to establish an accurate prediction model for fluid control performance and optimize control parameters; the hierarchical fluid parameter control strategy proposed in the present invention achieves coordinated optimization of the three control layers of pressure, temperature, and flow through multi-level calculations, thereby improving control accuracy; the hierarchical optimization method is used for inter-layer coordination, which effectively solves the problem of multi-parameter coupling and ensures the stability of the control system.

[0037] Reference Figure 3 In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0038] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0039] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0040] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0041] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0042] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0043] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A fluid control intelligent regulation method for a microfluidic chip, characterized in that: The method comprises: Based on the microfluidic chip, data is collected from the adaptive fluid controller, microfluidic channel pressure sensor array, temperature sensor array and flow sensor array to obtain real-time fluid parameter monitoring data; Inputting the real-time fluid parameter monitoring data into a dynamic self-learning control network model for feature extraction to obtain a multi-dimensional fluid feature matrix; Establish a three-dimensional fluid model of the microfluidic channel within a set pressure range and temperature range, and solve to obtain the flow field distribution data; The flow field distribution data is input into a Kriging proxy model, a mapping model between channel width, depth, length and fluid control performance is established, and the optimal geometric parameter combination of the microfluidic channel is calculated by a non-dominated sorting genetic algorithm; According to the multi-dimensional fluid characteristic matrix and the optimal geometric parameter combination, multi-level calculations are performed on fluid pressure uniformity, temperature distribution uniformity and flow stability to obtain a hierarchical fluid parameter control strategy.

2. The fluid control intelligent regulation method of the microfluidic chip according to claim 1, characterized in that: The method of collecting data from the adaptive fluid controller, the microfluidic channel pressure sensor array, the temperature sensor array and the flow sensor array based on the microfluidic chip to obtain real-time fluid parameter monitoring data includes: The microfluidic channel is gridded, and at the same time, the pressure sensor array is arranged at the grid nodes to obtain a pressure monitoring point distribution network; the key points of the inlet and outlet, corners and diversion nodes of the microfluidic channel are calibrated, and the temperature sensor array is arranged at the calibration point position to collect the temperature monitoring node matrix data; the flow monitoring points are set at the feed end and the discharge end of the microfluidic channel, and the flow sensor array is arranged at the monitoring point to collect the fluid flow parameters; Intelligently partitioning the reference fluid chamber based on the adaptive fluid controller to obtain a reference pressure monitoring point array, and performing real-time difference calculation between the pressure data of the pressure monitoring point distribution network and the pressure data of the reference pressure monitoring point array to obtain a pressure gradient distribution sequence; Based on the microfluidic chip, the pressure gradient distribution sequence, the temperature monitoring node matrix data and the fluid flow parameters are fused to obtain real-time fluid parameter monitoring data.

3. The fluid control intelligent regulation method of the microfluidic chip according to claim 2, characterized in that: The real-time fluid parameter monitoring data is input into a dynamic self-learning control network model for feature extraction to obtain a multi-dimensional fluid feature matrix, including: Inputting the pressure gradient distribution sequence of the real-time fluid parameter monitoring data into the adaptive fluid control module in the dynamic self-learning control network model for processing to obtain a pressure control feature sequence, wherein the adaptive fluid control module comprises a pressure feature extraction network and a pressure parameter optimization network; the pressure feature extraction network comprises three convolutional layers and two fully connected layers, and the pressure parameter optimization network comprises two fully connected layers and one output layer; Input the temperature monitoring node matrix data of the real-time fluid parameter monitoring data into the spatiotemporal feature extraction module in the dynamic self-learning control network model for three-dimensional convolution operation to obtain a temperature distribution feature sequence, wherein the spatiotemporal feature extraction module comprises a time series feature network and a space feature network, wherein the time series feature network adopts a long short-term memory network structure, and the space feature network adopts a three-dimensional convolution structure; Inputting the fluid flow parameters of the real-time fluid parameter monitoring data into the parameter self-adjustment module in the dynamic self-learning control network model for dynamic weight adjustment to obtain a flow change characteristic sequence, wherein the parameter self-adjustment module includes a flow prediction network and a parameter correction network; The pressure control feature sequence, the temperature distribution feature sequence and the flow change feature sequence are feature fused to obtain a fused feature matrix, and the fused feature matrix is ​​time-series reconstructed to obtain a multi-dimensional fluid feature matrix.

4. The fluid control intelligent regulation method of the microfluidic chip according to claim 3 is characterized in that: The three-dimensional fluid model of the microfluidic channel is established within the set pressure range and temperature range, and the flow field distribution data is obtained by solving the solution, including: Determine the pressure range and temperature range of the microfluidic channel, and perform three-dimensional solid modeling on the microfluidic channel to obtain a three-dimensional fluid model including a feed end, a discharge end, a corner, and a flow diversion node; Performing hexahedral structured grid division on the three-dimensional fluid model, and performing grid encryption on the inlet and outlet, corners and flow diversion nodes to obtain a computational domain grid; The mass conservation equation, momentum conservation equation and energy conservation equation are discretized on the computational domain grid using the finite volume method, and the second-order upwind scheme is used for spatial discretization to obtain a discrete control equation group; Solving the discrete control equations based on the SIMPLE algorithm to obtain initial flow field values, and encrypting and coarsening the number of grids based on the initial flow field values ​​to obtain an optimal grid distribution; The initial value of the flow field is iteratively solved on the optimal grid distribution to obtain a steady-state flow field solution, and based on the steady-state flow field solution, fluid pressure field, temperature field and velocity field data are extracted to obtain flow field distribution data.

5. The fluid control intelligent regulation method of the microfluidic chip according to claim 4, characterized in that: The flow field distribution data is input into the Kriging proxy model, a mapping model of channel width, depth, length and fluid control performance is established, and the optimal geometric parameter combination of the microfluidic channel is calculated by a non-dominated sorting genetic algorithm, including: Using the flow field distribution data as sample points, Latin hypercube sampling is performed on the width range, depth range and length range of the microfluidic channel to obtain a training sample space; A Kriging proxy model is established based on the training sample space, a Gaussian correlation function is used as a kernel function, a correlation parameter θ is set to 0.1, and the order of the regression function is set to 2, so as to obtain a correlation matrix of geometric parameters, and a maximum likelihood estimation function is constructed based on the correlation matrix, and hyperparameters are optimized to obtain the optimal interpolation weight of the Kriging model; Substituting the optimal interpolation weight into the Kriging proxy model, a mapping model between channel width, depth, length and pressure uniformity, temperature uniformity, and flow stability is established; Based on the mapping model, a non-dominated sorting genetic algorithm is optimized to obtain an evolved population, and a Pareto optimal solution is calculated based on the evolved population. Individuals are sorted using crowding distance, and individuals with a non-dominated level of 1 are selected to form an elite population to obtain a Pareto frontier. Normalizing the solution set in the Pareto front according to pressure uniformity, temperature uniformity and flow stability to obtain a candidate parameter combination; A sensitivity analysis is performed on the candidate parameter combinations, and the parameter combination with the best comprehensive evaluation value is selected as the optimal geometric parameter combination of the microfluidic channel.

6. The fluid control intelligent regulation method of the microfluidic chip according to claim 5, characterized in that: The Kriging proxy model is established based on the training sample space, the Gaussian correlation function is used as the kernel function, the correlation parameter θ is set to 0.1, the regression function order is 2, the correlation matrix of the geometric parameters is obtained, and the maximum likelihood estimation function is constructed based on the correlation matrix, and the hyperparameters are optimized to obtain the optimal interpolation weight of the Kriging model, including: Constructing a regression function matrix based on the training sample space, and expanding it using a second-order polynomial basis function to obtain regression model coefficients; Substituting the regression model coefficients into the Gaussian correlation function, calculating the spatial correlation between the sample points to obtain a correlation function matrix, and performing feature dimensionality reduction decomposition on the correlation function matrix to obtain a reduced-dimensional feature space; Constructing a maximum likelihood estimation function based on the reduced-dimensional feature space, using a logarithmic transformation form to obtain an optimization objective function, and inputting the optimization objective function into a pattern search algorithm to obtain an optimal hyperparameter; The optimal hyperparameters are cross-validated to obtain a model accuracy evaluation index, and the Kriging proxy model is modified according to the model accuracy evaluation index. The optimal interpolation weight of the Kriging model is obtained by calculating through the generalized least squares method.

7. The fluid control intelligent regulation method of the microfluidic chip according to claim 6, characterized in that: The multi-level calculation of fluid pressure uniformity, temperature distribution uniformity and flow stability is performed according to the multi-dimensional fluid characteristic matrix and the optimal geometric parameter combination to obtain a hierarchical fluid parameter control strategy, including: Performing correlation analysis on the pressure control feature sequence, the temperature distribution feature sequence and the flow change feature sequence in the multidimensional fluid feature matrix and the optimal geometric parameter combination, respectively, to obtain a feature-parameter mapping relationship; Input the feature-parameter mapping relationship into a control strategy generation module to construct an initial hierarchical control framework of a pressure control layer, a temperature control layer, and a flow control layer; Based on the initial hierarchical control framework, dynamic programming is performed on the pressure control layer data to obtain pressure uniformity control parameters; Based on the initial hierarchical control framework, model predictive control is performed on the temperature control layer data to obtain temperature distribution uniformity control parameters; Based on the initial hierarchical control framework, robust control calculation is performed on the flow control layer data to obtain flow stability control parameters; The pressure uniformity control parameter, the temperature distribution uniformity control parameter and the flow stability control parameter are input into a hierarchical coordinator, and hierarchical optimization is used to perform inter-layer coordination to obtain a hierarchical fluid parameter control strategy.

8. A fluid control intelligent regulation system for a microfluidic chip, characterized in that: The fluid control intelligent regulation method for a microfluidic chip according to any one of claims 1 to 7, wherein the fluid control intelligent regulation system for the microfluidic chip comprises: A data acquisition module is used to collect data from the adaptive fluid controller, the microfluidic channel pressure sensor array, the temperature sensor array and the flow sensor array based on the microfluidic chip to obtain real-time fluid parameter monitoring data; A feature extraction module, used for inputting the real-time fluid parameter monitoring data into a dynamic self-learning control network model for feature extraction to obtain a multi-dimensional fluid feature matrix; A solution module is used to establish a three-dimensional fluid model of the microfluidic channel within a set pressure range and temperature range, and solve to obtain flow field distribution data; Establishing a module for inputting the flow field distribution data into a Kriging proxy model, establishing a mapping model between channel width, depth, length and fluid control performance, and calculating the optimal geometric parameter combination of the microfluidic channel through a non-dominated sorting genetic algorithm; The calculation module is used to perform multi-level calculations on fluid pressure uniformity, temperature distribution uniformity and flow stability according to the multi-dimensional fluid characteristic matrix and the optimal geometric parameter combination, so as to obtain a hierarchical fluid parameter control strategy.

9. A computer device, characterized in that: It comprises a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and is characterized in that when the processor executes the computer program, the fluid control intelligent regulation method of the microfluidic chip described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the processor is enabled to execute the fluid control intelligent regulation method of a microfluidic chip according to any one of claims 1 to 7.

Images